Re: [RFC 1/5] user namespaces: Add a user_namespace as creator/owner of uts_namespace

[Greg KH <greg@kroah.com>]

On Fri, Dec 17, 2010 at 10:32:52PM -0800, Eric W. Biederman wrote:
> Greg KH <greg@kroah.com> writes:
> 
> > On Fri, Dec 17, 2010 at 12:40:11PM -0800, Eric W. Biederman wrote:
> >> Greg KH <greg@kroah.com> writes:
> >> 
> >> >> There are also major benefits to have the version of something that is
> >> >> never freed never going away, because it means you can just reference it
> >> >> in code.  So while I would be happy to say this is special don't use a
> >> >> kref and roll the reference counting logic by hand, we aren't
> >> >> dynamically allocating init_uts_ns any time soon.
> >> >
> >> > Why have a reference count at all if it's not needed or used here?
> >> 
> >> We have to reference count every other uts namespace.
> >
> > Ok, that makes sense, then also please dynamically create this one, do
> > not create a static kref.
> 
> Nope.  It's a bad idea.  It messes up the kernel bootstrap if you do
> that, and it makes this one structure different from every other
> structure init_task uses.

{sigh}

Ok, but I really don't like this use.

Also, don't go messing with that ATOMIC_INIT() to be a higher value, as
this patch series did, as that really implies that it is being used
incorrectly, right?

thanks,

greg k-h